Spatially-Adaptive Filter Units for Compact and Efficient Deep Neural Networks
Abstract Convolutional neural networks excel in a number of computer vision tasks. One of their most crucial architectural elements is the effective receptive field size, which has to be manually set to accommodate a specific task. Standard solutions involve large kernels, down/up-sampling and dilat...
Ausführliche Beschreibung
Gespeichert in:
| Autor*in: |
Tabernik, Domen [verfasserIn] |
|---|
| Format: |
Artikel |
|---|---|
| Sprache: |
Englisch |
| Erschienen: |
2020 |
|---|
| Schlagwörter: |
|---|
| Anmerkung: |
© Springer Science+Business Media, LLC, part of Springer Nature 2020 |
|---|
| Übergeordnetes Werk: |
Enthalten in: International journal of computer vision - Springer US, 1987, 128(2020), 8-9 vom: 02. Jan., Seite 2049-2067 |
|---|---|
| Übergeordnetes Werk: |
volume:128 ; year:2020 ; number:8-9 ; day:02 ; month:01 ; pages:2049-2067 |
| Links: |
|---|
| DOI / URN: |
10.1007/s11263-019-01282-1 |
|---|
| Katalog-ID: |
OLC211889029X |
|---|
| LEADER | 01000naa a22002652 4500 | ||
|---|---|---|---|
| 001 | OLC211889029X | ||
| 003 | DE-627 | ||
| 005 | 20230504162839.0 | ||
| 007 | tu | ||
| 008 | 230504s2020 xx ||||| 00| ||eng c | ||
| 024 | 7 | |a 10.1007/s11263-019-01282-1 |2 doi | |
| 035 | |a (DE-627)OLC211889029X | ||
| 035 | |a (DE-He213)s11263-019-01282-1-p | ||
| 040 | |a DE-627 |b ger |c DE-627 |e rakwb | ||
| 041 | |a eng | ||
| 082 | 0 | 4 | |a 004 |q VZ |
| 100 | 1 | |a Tabernik, Domen |e verfasserin |0 (orcid)0000-0002-5613-5882 |4 aut | |
| 245 | 1 | 0 | |a Spatially-Adaptive Filter Units for Compact and Efficient Deep Neural Networks |
| 264 | 1 | |c 2020 | |
| 336 | |a Text |b txt |2 rdacontent | ||
| 337 | |a ohne Hilfsmittel zu benutzen |b n |2 rdamedia | ||
| 338 | |a Band |b nc |2 rdacarrier | ||
| 500 | |a © Springer Science+Business Media, LLC, part of Springer Nature 2020 | ||
| 520 | |a Abstract Convolutional neural networks excel in a number of computer vision tasks. One of their most crucial architectural elements is the effective receptive field size, which has to be manually set to accommodate a specific task. Standard solutions involve large kernels, down/up-sampling and dilated convolutions. These require testing a variety of dilation and down/up-sampling factors and result in non-compact networks and large number of parameters. We address this issue by proposing a new convolution filter composed of displaced aggregation units (DAU). DAUs learn spatial displacements and adapt the receptive field sizes of individual convolution filters to a given problem, thus reducing the need for hand-crafted modifications. DAUs provide a seamless substitution of convolutional filters in existing state-of-the-art architectures, which we demonstrate on AlexNet, ResNet50, ResNet101, DeepLab and SRN-DeblurNet. The benefits of this design are demonstrated on a variety of computer vision tasks and datasets, such as image classification (ILSVRC 2012), semantic segmentation (PASCAL VOC 2011, Cityscape) and blind image de-blurring (GOPRO). Results show that DAUs efficiently allocate parameters resulting in up to 4$$\times $$ more compact networks in terms of the number of parameters at similar or better performance. | ||
| 650 | 4 | |a Compact ConvNets | |
| 650 | 4 | |a Efficient ConvNets | |
| 650 | 4 | |a Displacement units | |
| 650 | 4 | |a Adjustable receptive fields | |
| 700 | 1 | |a Kristan, Matej |4 aut | |
| 700 | 1 | |a Leonardis, Aleš |4 aut | |
| 773 | 0 | 8 | |i Enthalten in |t International journal of computer vision |d Springer US, 1987 |g 128(2020), 8-9 vom: 02. Jan., Seite 2049-2067 |w (DE-627)129354252 |w (DE-600)155895-X |w (DE-576)018081428 |x 0920-5691 |7 nnns |
| 773 | 1 | 8 | |g volume:128 |g year:2020 |g number:8-9 |g day:02 |g month:01 |g pages:2049-2067 |
| 856 | 4 | 1 | |u https://doi.org/10.1007/s11263-019-01282-1 |z lizenzpflichtig |3 Volltext |
| 912 | |a GBV_USEFLAG_A | ||
| 912 | |a SYSFLAG_A | ||
| 912 | |a GBV_OLC | ||
| 912 | |a SSG-OLC-MAT | ||
| 912 | |a GBV_ILN_2244 | ||
| 951 | |a AR | ||
| 952 | |d 128 |j 2020 |e 8-9 |b 02 |c 01 |h 2049-2067 | ||
| author_variant |
d t dt m k mk a l al |
|---|---|
| matchkey_str |
article:09205691:2020----::ptalaatvfleuisocmatnefcet |
| hierarchy_sort_str |
2020 |
| publishDate |
2020 |
| allfields |
10.1007/s11263-019-01282-1 doi (DE-627)OLC211889029X (DE-He213)s11263-019-01282-1-p DE-627 ger DE-627 rakwb eng 004 VZ Tabernik, Domen verfasserin (orcid)0000-0002-5613-5882 aut Spatially-Adaptive Filter Units for Compact and Efficient Deep Neural Networks 2020 Text txt rdacontent ohne Hilfsmittel zu benutzen n rdamedia Band nc rdacarrier © Springer Science+Business Media, LLC, part of Springer Nature 2020 Abstract Convolutional neural networks excel in a number of computer vision tasks. One of their most crucial architectural elements is the effective receptive field size, which has to be manually set to accommodate a specific task. Standard solutions involve large kernels, down/up-sampling and dilated convolutions. These require testing a variety of dilation and down/up-sampling factors and result in non-compact networks and large number of parameters. We address this issue by proposing a new convolution filter composed of displaced aggregation units (DAU). DAUs learn spatial displacements and adapt the receptive field sizes of individual convolution filters to a given problem, thus reducing the need for hand-crafted modifications. DAUs provide a seamless substitution of convolutional filters in existing state-of-the-art architectures, which we demonstrate on AlexNet, ResNet50, ResNet101, DeepLab and SRN-DeblurNet. The benefits of this design are demonstrated on a variety of computer vision tasks and datasets, such as image classification (ILSVRC 2012), semantic segmentation (PASCAL VOC 2011, Cityscape) and blind image de-blurring (GOPRO). Results show that DAUs efficiently allocate parameters resulting in up to 4$$\times $$ more compact networks in terms of the number of parameters at similar or better performance. Compact ConvNets Efficient ConvNets Displacement units Adjustable receptive fields Kristan, Matej aut Leonardis, Aleš aut Enthalten in International journal of computer vision Springer US, 1987 128(2020), 8-9 vom: 02. Jan., Seite 2049-2067 (DE-627)129354252 (DE-600)155895-X (DE-576)018081428 0920-5691 nnns volume:128 year:2020 number:8-9 day:02 month:01 pages:2049-2067 https://doi.org/10.1007/s11263-019-01282-1 lizenzpflichtig Volltext GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-MAT GBV_ILN_2244 AR 128 2020 8-9 02 01 2049-2067 |
| spelling |
10.1007/s11263-019-01282-1 doi (DE-627)OLC211889029X (DE-He213)s11263-019-01282-1-p DE-627 ger DE-627 rakwb eng 004 VZ Tabernik, Domen verfasserin (orcid)0000-0002-5613-5882 aut Spatially-Adaptive Filter Units for Compact and Efficient Deep Neural Networks 2020 Text txt rdacontent ohne Hilfsmittel zu benutzen n rdamedia Band nc rdacarrier © Springer Science+Business Media, LLC, part of Springer Nature 2020 Abstract Convolutional neural networks excel in a number of computer vision tasks. One of their most crucial architectural elements is the effective receptive field size, which has to be manually set to accommodate a specific task. Standard solutions involve large kernels, down/up-sampling and dilated convolutions. These require testing a variety of dilation and down/up-sampling factors and result in non-compact networks and large number of parameters. We address this issue by proposing a new convolution filter composed of displaced aggregation units (DAU). DAUs learn spatial displacements and adapt the receptive field sizes of individual convolution filters to a given problem, thus reducing the need for hand-crafted modifications. DAUs provide a seamless substitution of convolutional filters in existing state-of-the-art architectures, which we demonstrate on AlexNet, ResNet50, ResNet101, DeepLab and SRN-DeblurNet. The benefits of this design are demonstrated on a variety of computer vision tasks and datasets, such as image classification (ILSVRC 2012), semantic segmentation (PASCAL VOC 2011, Cityscape) and blind image de-blurring (GOPRO). Results show that DAUs efficiently allocate parameters resulting in up to 4$$\times $$ more compact networks in terms of the number of parameters at similar or better performance. Compact ConvNets Efficient ConvNets Displacement units Adjustable receptive fields Kristan, Matej aut Leonardis, Aleš aut Enthalten in International journal of computer vision Springer US, 1987 128(2020), 8-9 vom: 02. Jan., Seite 2049-2067 (DE-627)129354252 (DE-600)155895-X (DE-576)018081428 0920-5691 nnns volume:128 year:2020 number:8-9 day:02 month:01 pages:2049-2067 https://doi.org/10.1007/s11263-019-01282-1 lizenzpflichtig Volltext GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-MAT GBV_ILN_2244 AR 128 2020 8-9 02 01 2049-2067 |
| allfields_unstemmed |
10.1007/s11263-019-01282-1 doi (DE-627)OLC211889029X (DE-He213)s11263-019-01282-1-p DE-627 ger DE-627 rakwb eng 004 VZ Tabernik, Domen verfasserin (orcid)0000-0002-5613-5882 aut Spatially-Adaptive Filter Units for Compact and Efficient Deep Neural Networks 2020 Text txt rdacontent ohne Hilfsmittel zu benutzen n rdamedia Band nc rdacarrier © Springer Science+Business Media, LLC, part of Springer Nature 2020 Abstract Convolutional neural networks excel in a number of computer vision tasks. One of their most crucial architectural elements is the effective receptive field size, which has to be manually set to accommodate a specific task. Standard solutions involve large kernels, down/up-sampling and dilated convolutions. These require testing a variety of dilation and down/up-sampling factors and result in non-compact networks and large number of parameters. We address this issue by proposing a new convolution filter composed of displaced aggregation units (DAU). DAUs learn spatial displacements and adapt the receptive field sizes of individual convolution filters to a given problem, thus reducing the need for hand-crafted modifications. DAUs provide a seamless substitution of convolutional filters in existing state-of-the-art architectures, which we demonstrate on AlexNet, ResNet50, ResNet101, DeepLab and SRN-DeblurNet. The benefits of this design are demonstrated on a variety of computer vision tasks and datasets, such as image classification (ILSVRC 2012), semantic segmentation (PASCAL VOC 2011, Cityscape) and blind image de-blurring (GOPRO). Results show that DAUs efficiently allocate parameters resulting in up to 4$$\times $$ more compact networks in terms of the number of parameters at similar or better performance. Compact ConvNets Efficient ConvNets Displacement units Adjustable receptive fields Kristan, Matej aut Leonardis, Aleš aut Enthalten in International journal of computer vision Springer US, 1987 128(2020), 8-9 vom: 02. Jan., Seite 2049-2067 (DE-627)129354252 (DE-600)155895-X (DE-576)018081428 0920-5691 nnns volume:128 year:2020 number:8-9 day:02 month:01 pages:2049-2067 https://doi.org/10.1007/s11263-019-01282-1 lizenzpflichtig Volltext GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-MAT GBV_ILN_2244 AR 128 2020 8-9 02 01 2049-2067 |
| allfieldsGer |
10.1007/s11263-019-01282-1 doi (DE-627)OLC211889029X (DE-He213)s11263-019-01282-1-p DE-627 ger DE-627 rakwb eng 004 VZ Tabernik, Domen verfasserin (orcid)0000-0002-5613-5882 aut Spatially-Adaptive Filter Units for Compact and Efficient Deep Neural Networks 2020 Text txt rdacontent ohne Hilfsmittel zu benutzen n rdamedia Band nc rdacarrier © Springer Science+Business Media, LLC, part of Springer Nature 2020 Abstract Convolutional neural networks excel in a number of computer vision tasks. One of their most crucial architectural elements is the effective receptive field size, which has to be manually set to accommodate a specific task. Standard solutions involve large kernels, down/up-sampling and dilated convolutions. These require testing a variety of dilation and down/up-sampling factors and result in non-compact networks and large number of parameters. We address this issue by proposing a new convolution filter composed of displaced aggregation units (DAU). DAUs learn spatial displacements and adapt the receptive field sizes of individual convolution filters to a given problem, thus reducing the need for hand-crafted modifications. DAUs provide a seamless substitution of convolutional filters in existing state-of-the-art architectures, which we demonstrate on AlexNet, ResNet50, ResNet101, DeepLab and SRN-DeblurNet. The benefits of this design are demonstrated on a variety of computer vision tasks and datasets, such as image classification (ILSVRC 2012), semantic segmentation (PASCAL VOC 2011, Cityscape) and blind image de-blurring (GOPRO). Results show that DAUs efficiently allocate parameters resulting in up to 4$$\times $$ more compact networks in terms of the number of parameters at similar or better performance. Compact ConvNets Efficient ConvNets Displacement units Adjustable receptive fields Kristan, Matej aut Leonardis, Aleš aut Enthalten in International journal of computer vision Springer US, 1987 128(2020), 8-9 vom: 02. Jan., Seite 2049-2067 (DE-627)129354252 (DE-600)155895-X (DE-576)018081428 0920-5691 nnns volume:128 year:2020 number:8-9 day:02 month:01 pages:2049-2067 https://doi.org/10.1007/s11263-019-01282-1 lizenzpflichtig Volltext GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-MAT GBV_ILN_2244 AR 128 2020 8-9 02 01 2049-2067 |
| allfieldsSound |
10.1007/s11263-019-01282-1 doi (DE-627)OLC211889029X (DE-He213)s11263-019-01282-1-p DE-627 ger DE-627 rakwb eng 004 VZ Tabernik, Domen verfasserin (orcid)0000-0002-5613-5882 aut Spatially-Adaptive Filter Units for Compact and Efficient Deep Neural Networks 2020 Text txt rdacontent ohne Hilfsmittel zu benutzen n rdamedia Band nc rdacarrier © Springer Science+Business Media, LLC, part of Springer Nature 2020 Abstract Convolutional neural networks excel in a number of computer vision tasks. One of their most crucial architectural elements is the effective receptive field size, which has to be manually set to accommodate a specific task. Standard solutions involve large kernels, down/up-sampling and dilated convolutions. These require testing a variety of dilation and down/up-sampling factors and result in non-compact networks and large number of parameters. We address this issue by proposing a new convolution filter composed of displaced aggregation units (DAU). DAUs learn spatial displacements and adapt the receptive field sizes of individual convolution filters to a given problem, thus reducing the need for hand-crafted modifications. DAUs provide a seamless substitution of convolutional filters in existing state-of-the-art architectures, which we demonstrate on AlexNet, ResNet50, ResNet101, DeepLab and SRN-DeblurNet. The benefits of this design are demonstrated on a variety of computer vision tasks and datasets, such as image classification (ILSVRC 2012), semantic segmentation (PASCAL VOC 2011, Cityscape) and blind image de-blurring (GOPRO). Results show that DAUs efficiently allocate parameters resulting in up to 4$$\times $$ more compact networks in terms of the number of parameters at similar or better performance. Compact ConvNets Efficient ConvNets Displacement units Adjustable receptive fields Kristan, Matej aut Leonardis, Aleš aut Enthalten in International journal of computer vision Springer US, 1987 128(2020), 8-9 vom: 02. Jan., Seite 2049-2067 (DE-627)129354252 (DE-600)155895-X (DE-576)018081428 0920-5691 nnns volume:128 year:2020 number:8-9 day:02 month:01 pages:2049-2067 https://doi.org/10.1007/s11263-019-01282-1 lizenzpflichtig Volltext GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-MAT GBV_ILN_2244 AR 128 2020 8-9 02 01 2049-2067 |
| language |
English |
| source |
Enthalten in International journal of computer vision 128(2020), 8-9 vom: 02. Jan., Seite 2049-2067 volume:128 year:2020 number:8-9 day:02 month:01 pages:2049-2067 |
| sourceStr |
Enthalten in International journal of computer vision 128(2020), 8-9 vom: 02. Jan., Seite 2049-2067 volume:128 year:2020 number:8-9 day:02 month:01 pages:2049-2067 |
| format_phy_str_mv |
Article |
| institution |
findex.gbv.de |
| topic_facet |
Compact ConvNets Efficient ConvNets Displacement units Adjustable receptive fields |
| dewey-raw |
004 |
| isfreeaccess_bool |
false |
| container_title |
International journal of computer vision |
| authorswithroles_txt_mv |
Tabernik, Domen @@aut@@ Kristan, Matej @@aut@@ Leonardis, Aleš @@aut@@ |
| publishDateDaySort_date |
2020-01-02T00:00:00Z |
| hierarchy_top_id |
129354252 |
| dewey-sort |
14 |
| id |
OLC211889029X |
| language_de |
englisch |
| fullrecord |
<?xml version="1.0" encoding="UTF-8"?><collection xmlns="http://www.loc.gov/MARC21/slim"><record><leader>01000naa a22002652 4500</leader><controlfield tag="001">OLC211889029X</controlfield><controlfield tag="003">DE-627</controlfield><controlfield tag="005">20230504162839.0</controlfield><controlfield tag="007">tu</controlfield><controlfield tag="008">230504s2020 xx ||||| 00| ||eng c</controlfield><datafield tag="024" ind1="7" ind2=" "><subfield code="a">10.1007/s11263-019-01282-1</subfield><subfield code="2">doi</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(DE-627)OLC211889029X</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(DE-He213)s11263-019-01282-1-p</subfield></datafield><datafield tag="040" ind1=" " ind2=" "><subfield code="a">DE-627</subfield><subfield code="b">ger</subfield><subfield code="c">DE-627</subfield><subfield code="e">rakwb</subfield></datafield><datafield tag="041" ind1=" " ind2=" "><subfield code="a">eng</subfield></datafield><datafield tag="082" ind1="0" ind2="4"><subfield code="a">004</subfield><subfield code="q">VZ</subfield></datafield><datafield tag="100" ind1="1" ind2=" "><subfield code="a">Tabernik, Domen</subfield><subfield code="e">verfasserin</subfield><subfield code="0">(orcid)0000-0002-5613-5882</subfield><subfield code="4">aut</subfield></datafield><datafield tag="245" ind1="1" ind2="0"><subfield code="a">Spatially-Adaptive Filter Units for Compact and Efficient Deep Neural Networks</subfield></datafield><datafield tag="264" ind1=" " ind2="1"><subfield code="c">2020</subfield></datafield><datafield tag="336" ind1=" " ind2=" "><subfield code="a">Text</subfield><subfield code="b">txt</subfield><subfield code="2">rdacontent</subfield></datafield><datafield tag="337" ind1=" " ind2=" "><subfield code="a">ohne Hilfsmittel zu benutzen</subfield><subfield code="b">n</subfield><subfield code="2">rdamedia</subfield></datafield><datafield tag="338" ind1=" " ind2=" "><subfield code="a">Band</subfield><subfield code="b">nc</subfield><subfield code="2">rdacarrier</subfield></datafield><datafield tag="500" ind1=" " ind2=" "><subfield code="a">© Springer Science+Business Media, LLC, part of Springer Nature 2020</subfield></datafield><datafield tag="520" ind1=" " ind2=" "><subfield code="a">Abstract Convolutional neural networks excel in a number of computer vision tasks. One of their most crucial architectural elements is the effective receptive field size, which has to be manually set to accommodate a specific task. Standard solutions involve large kernels, down/up-sampling and dilated convolutions. These require testing a variety of dilation and down/up-sampling factors and result in non-compact networks and large number of parameters. We address this issue by proposing a new convolution filter composed of displaced aggregation units (DAU). DAUs learn spatial displacements and adapt the receptive field sizes of individual convolution filters to a given problem, thus reducing the need for hand-crafted modifications. DAUs provide a seamless substitution of convolutional filters in existing state-of-the-art architectures, which we demonstrate on AlexNet, ResNet50, ResNet101, DeepLab and SRN-DeblurNet. The benefits of this design are demonstrated on a variety of computer vision tasks and datasets, such as image classification (ILSVRC 2012), semantic segmentation (PASCAL VOC 2011, Cityscape) and blind image de-blurring (GOPRO). Results show that DAUs efficiently allocate parameters resulting in up to 4$$\times $$ more compact networks in terms of the number of parameters at similar or better performance.</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Compact ConvNets</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Efficient ConvNets</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Displacement units</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Adjustable receptive fields</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Kristan, Matej</subfield><subfield code="4">aut</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Leonardis, Aleš</subfield><subfield code="4">aut</subfield></datafield><datafield tag="773" ind1="0" ind2="8"><subfield code="i">Enthalten in</subfield><subfield code="t">International journal of computer vision</subfield><subfield code="d">Springer US, 1987</subfield><subfield code="g">128(2020), 8-9 vom: 02. Jan., Seite 2049-2067</subfield><subfield code="w">(DE-627)129354252</subfield><subfield code="w">(DE-600)155895-X</subfield><subfield code="w">(DE-576)018081428</subfield><subfield code="x">0920-5691</subfield><subfield code="7">nnns</subfield></datafield><datafield tag="773" ind1="1" ind2="8"><subfield code="g">volume:128</subfield><subfield code="g">year:2020</subfield><subfield code="g">number:8-9</subfield><subfield code="g">day:02</subfield><subfield code="g">month:01</subfield><subfield code="g">pages:2049-2067</subfield></datafield><datafield tag="856" ind1="4" ind2="1"><subfield code="u">https://doi.org/10.1007/s11263-019-01282-1</subfield><subfield code="z">lizenzpflichtig</subfield><subfield code="3">Volltext</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_USEFLAG_A</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SYSFLAG_A</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_OLC</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SSG-OLC-MAT</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_ILN_2244</subfield></datafield><datafield tag="951" ind1=" " ind2=" "><subfield code="a">AR</subfield></datafield><datafield tag="952" ind1=" " ind2=" "><subfield code="d">128</subfield><subfield code="j">2020</subfield><subfield code="e">8-9</subfield><subfield code="b">02</subfield><subfield code="c">01</subfield><subfield code="h">2049-2067</subfield></datafield></record></collection>
|
| author |
Tabernik, Domen |
| spellingShingle |
Tabernik, Domen ddc 004 misc Compact ConvNets misc Efficient ConvNets misc Displacement units misc Adjustable receptive fields Spatially-Adaptive Filter Units for Compact and Efficient Deep Neural Networks |
| authorStr |
Tabernik, Domen |
| ppnlink_with_tag_str_mv |
@@773@@(DE-627)129354252 |
| format |
Article |
| dewey-ones |
004 - Data processing & computer science |
| delete_txt_mv |
keep |
| author_role |
aut aut aut |
| collection |
OLC |
| remote_str |
false |
| illustrated |
Not Illustrated |
| issn |
0920-5691 |
| topic_title |
004 VZ Spatially-Adaptive Filter Units for Compact and Efficient Deep Neural Networks Compact ConvNets Efficient ConvNets Displacement units Adjustable receptive fields |
| topic |
ddc 004 misc Compact ConvNets misc Efficient ConvNets misc Displacement units misc Adjustable receptive fields |
| topic_unstemmed |
ddc 004 misc Compact ConvNets misc Efficient ConvNets misc Displacement units misc Adjustable receptive fields |
| topic_browse |
ddc 004 misc Compact ConvNets misc Efficient ConvNets misc Displacement units misc Adjustable receptive fields |
| format_facet |
Aufsätze Gedruckte Aufsätze |
| format_main_str_mv |
Text Zeitschrift/Artikel |
| carriertype_str_mv |
nc |
| hierarchy_parent_title |
International journal of computer vision |
| hierarchy_parent_id |
129354252 |
| dewey-tens |
000 - Computer science, knowledge & systems |
| hierarchy_top_title |
International journal of computer vision |
| isfreeaccess_txt |
false |
| familylinks_str_mv |
(DE-627)129354252 (DE-600)155895-X (DE-576)018081428 |
| title |
Spatially-Adaptive Filter Units for Compact and Efficient Deep Neural Networks |
| ctrlnum |
(DE-627)OLC211889029X (DE-He213)s11263-019-01282-1-p |
| title_full |
Spatially-Adaptive Filter Units for Compact and Efficient Deep Neural Networks |
| author_sort |
Tabernik, Domen |
| journal |
International journal of computer vision |
| journalStr |
International journal of computer vision |
| lang_code |
eng |
| isOA_bool |
false |
| dewey-hundreds |
000 - Computer science, information & general works |
| recordtype |
marc |
| publishDateSort |
2020 |
| contenttype_str_mv |
txt |
| container_start_page |
2049 |
| author_browse |
Tabernik, Domen Kristan, Matej Leonardis, Aleš |
| container_volume |
128 |
| class |
004 VZ |
| format_se |
Aufsätze |
| author-letter |
Tabernik, Domen |
| doi_str_mv |
10.1007/s11263-019-01282-1 |
| normlink |
(ORCID)0000-0002-5613-5882 |
| normlink_prefix_str_mv |
(orcid)0000-0002-5613-5882 |
| dewey-full |
004 |
| title_sort |
spatially-adaptive filter units for compact and efficient deep neural networks |
| title_auth |
Spatially-Adaptive Filter Units for Compact and Efficient Deep Neural Networks |
| abstract |
Abstract Convolutional neural networks excel in a number of computer vision tasks. One of their most crucial architectural elements is the effective receptive field size, which has to be manually set to accommodate a specific task. Standard solutions involve large kernels, down/up-sampling and dilated convolutions. These require testing a variety of dilation and down/up-sampling factors and result in non-compact networks and large number of parameters. We address this issue by proposing a new convolution filter composed of displaced aggregation units (DAU). DAUs learn spatial displacements and adapt the receptive field sizes of individual convolution filters to a given problem, thus reducing the need for hand-crafted modifications. DAUs provide a seamless substitution of convolutional filters in existing state-of-the-art architectures, which we demonstrate on AlexNet, ResNet50, ResNet101, DeepLab and SRN-DeblurNet. The benefits of this design are demonstrated on a variety of computer vision tasks and datasets, such as image classification (ILSVRC 2012), semantic segmentation (PASCAL VOC 2011, Cityscape) and blind image de-blurring (GOPRO). Results show that DAUs efficiently allocate parameters resulting in up to 4$$\times $$ more compact networks in terms of the number of parameters at similar or better performance. © Springer Science+Business Media, LLC, part of Springer Nature 2020 |
| abstractGer |
Abstract Convolutional neural networks excel in a number of computer vision tasks. One of their most crucial architectural elements is the effective receptive field size, which has to be manually set to accommodate a specific task. Standard solutions involve large kernels, down/up-sampling and dilated convolutions. These require testing a variety of dilation and down/up-sampling factors and result in non-compact networks and large number of parameters. We address this issue by proposing a new convolution filter composed of displaced aggregation units (DAU). DAUs learn spatial displacements and adapt the receptive field sizes of individual convolution filters to a given problem, thus reducing the need for hand-crafted modifications. DAUs provide a seamless substitution of convolutional filters in existing state-of-the-art architectures, which we demonstrate on AlexNet, ResNet50, ResNet101, DeepLab and SRN-DeblurNet. The benefits of this design are demonstrated on a variety of computer vision tasks and datasets, such as image classification (ILSVRC 2012), semantic segmentation (PASCAL VOC 2011, Cityscape) and blind image de-blurring (GOPRO). Results show that DAUs efficiently allocate parameters resulting in up to 4$$\times $$ more compact networks in terms of the number of parameters at similar or better performance. © Springer Science+Business Media, LLC, part of Springer Nature 2020 |
| abstract_unstemmed |
Abstract Convolutional neural networks excel in a number of computer vision tasks. One of their most crucial architectural elements is the effective receptive field size, which has to be manually set to accommodate a specific task. Standard solutions involve large kernels, down/up-sampling and dilated convolutions. These require testing a variety of dilation and down/up-sampling factors and result in non-compact networks and large number of parameters. We address this issue by proposing a new convolution filter composed of displaced aggregation units (DAU). DAUs learn spatial displacements and adapt the receptive field sizes of individual convolution filters to a given problem, thus reducing the need for hand-crafted modifications. DAUs provide a seamless substitution of convolutional filters in existing state-of-the-art architectures, which we demonstrate on AlexNet, ResNet50, ResNet101, DeepLab and SRN-DeblurNet. The benefits of this design are demonstrated on a variety of computer vision tasks and datasets, such as image classification (ILSVRC 2012), semantic segmentation (PASCAL VOC 2011, Cityscape) and blind image de-blurring (GOPRO). Results show that DAUs efficiently allocate parameters resulting in up to 4$$\times $$ more compact networks in terms of the number of parameters at similar or better performance. © Springer Science+Business Media, LLC, part of Springer Nature 2020 |
| collection_details |
GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-MAT GBV_ILN_2244 |
| container_issue |
8-9 |
| title_short |
Spatially-Adaptive Filter Units for Compact and Efficient Deep Neural Networks |
| url |
https://doi.org/10.1007/s11263-019-01282-1 |
| remote_bool |
false |
| author2 |
Kristan, Matej Leonardis, Aleš |
| author2Str |
Kristan, Matej Leonardis, Aleš |
| ppnlink |
129354252 |
| mediatype_str_mv |
n |
| isOA_txt |
false |
| hochschulschrift_bool |
false |
| doi_str |
10.1007/s11263-019-01282-1 |
| up_date |
2024-07-03T22:05:27.976Z |
| _version_ |
1803597201254907904 |
| fullrecord_marcxml |
<?xml version="1.0" encoding="UTF-8"?><collection xmlns="http://www.loc.gov/MARC21/slim"><record><leader>01000naa a22002652 4500</leader><controlfield tag="001">OLC211889029X</controlfield><controlfield tag="003">DE-627</controlfield><controlfield tag="005">20230504162839.0</controlfield><controlfield tag="007">tu</controlfield><controlfield tag="008">230504s2020 xx ||||| 00| ||eng c</controlfield><datafield tag="024" ind1="7" ind2=" "><subfield code="a">10.1007/s11263-019-01282-1</subfield><subfield code="2">doi</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(DE-627)OLC211889029X</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(DE-He213)s11263-019-01282-1-p</subfield></datafield><datafield tag="040" ind1=" " ind2=" "><subfield code="a">DE-627</subfield><subfield code="b">ger</subfield><subfield code="c">DE-627</subfield><subfield code="e">rakwb</subfield></datafield><datafield tag="041" ind1=" " ind2=" "><subfield code="a">eng</subfield></datafield><datafield tag="082" ind1="0" ind2="4"><subfield code="a">004</subfield><subfield code="q">VZ</subfield></datafield><datafield tag="100" ind1="1" ind2=" "><subfield code="a">Tabernik, Domen</subfield><subfield code="e">verfasserin</subfield><subfield code="0">(orcid)0000-0002-5613-5882</subfield><subfield code="4">aut</subfield></datafield><datafield tag="245" ind1="1" ind2="0"><subfield code="a">Spatially-Adaptive Filter Units for Compact and Efficient Deep Neural Networks</subfield></datafield><datafield tag="264" ind1=" " ind2="1"><subfield code="c">2020</subfield></datafield><datafield tag="336" ind1=" " ind2=" "><subfield code="a">Text</subfield><subfield code="b">txt</subfield><subfield code="2">rdacontent</subfield></datafield><datafield tag="337" ind1=" " ind2=" "><subfield code="a">ohne Hilfsmittel zu benutzen</subfield><subfield code="b">n</subfield><subfield code="2">rdamedia</subfield></datafield><datafield tag="338" ind1=" " ind2=" "><subfield code="a">Band</subfield><subfield code="b">nc</subfield><subfield code="2">rdacarrier</subfield></datafield><datafield tag="500" ind1=" " ind2=" "><subfield code="a">© Springer Science+Business Media, LLC, part of Springer Nature 2020</subfield></datafield><datafield tag="520" ind1=" " ind2=" "><subfield code="a">Abstract Convolutional neural networks excel in a number of computer vision tasks. One of their most crucial architectural elements is the effective receptive field size, which has to be manually set to accommodate a specific task. Standard solutions involve large kernels, down/up-sampling and dilated convolutions. These require testing a variety of dilation and down/up-sampling factors and result in non-compact networks and large number of parameters. We address this issue by proposing a new convolution filter composed of displaced aggregation units (DAU). DAUs learn spatial displacements and adapt the receptive field sizes of individual convolution filters to a given problem, thus reducing the need for hand-crafted modifications. DAUs provide a seamless substitution of convolutional filters in existing state-of-the-art architectures, which we demonstrate on AlexNet, ResNet50, ResNet101, DeepLab and SRN-DeblurNet. The benefits of this design are demonstrated on a variety of computer vision tasks and datasets, such as image classification (ILSVRC 2012), semantic segmentation (PASCAL VOC 2011, Cityscape) and blind image de-blurring (GOPRO). Results show that DAUs efficiently allocate parameters resulting in up to 4$$\times $$ more compact networks in terms of the number of parameters at similar or better performance.</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Compact ConvNets</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Efficient ConvNets</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Displacement units</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Adjustable receptive fields</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Kristan, Matej</subfield><subfield code="4">aut</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Leonardis, Aleš</subfield><subfield code="4">aut</subfield></datafield><datafield tag="773" ind1="0" ind2="8"><subfield code="i">Enthalten in</subfield><subfield code="t">International journal of computer vision</subfield><subfield code="d">Springer US, 1987</subfield><subfield code="g">128(2020), 8-9 vom: 02. Jan., Seite 2049-2067</subfield><subfield code="w">(DE-627)129354252</subfield><subfield code="w">(DE-600)155895-X</subfield><subfield code="w">(DE-576)018081428</subfield><subfield code="x">0920-5691</subfield><subfield code="7">nnns</subfield></datafield><datafield tag="773" ind1="1" ind2="8"><subfield code="g">volume:128</subfield><subfield code="g">year:2020</subfield><subfield code="g">number:8-9</subfield><subfield code="g">day:02</subfield><subfield code="g">month:01</subfield><subfield code="g">pages:2049-2067</subfield></datafield><datafield tag="856" ind1="4" ind2="1"><subfield code="u">https://doi.org/10.1007/s11263-019-01282-1</subfield><subfield code="z">lizenzpflichtig</subfield><subfield code="3">Volltext</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_USEFLAG_A</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SYSFLAG_A</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_OLC</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SSG-OLC-MAT</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_ILN_2244</subfield></datafield><datafield tag="951" ind1=" " ind2=" "><subfield code="a">AR</subfield></datafield><datafield tag="952" ind1=" " ind2=" "><subfield code="d">128</subfield><subfield code="j">2020</subfield><subfield code="e">8-9</subfield><subfield code="b">02</subfield><subfield code="c">01</subfield><subfield code="h">2049-2067</subfield></datafield></record></collection>
|
| score |
7.3984203 |